DeepSpeech

An open-source speech-to-text engine utilizing Baidu's Deep Speech architecture and implemented via TensorFlow.

DeepSpeech transforms audio into text using a production-ready model trained on thousands of hours of voice data. It leverages a recurrent neural network (RNN) to map spectrograms directly to character sequences, bypassing traditional phonetic engineering. Developed by Mozilla, the engine supports real-time transcription on hardware ranging from Raspberry Pi 4 devices to high-end NVIDIA GPUs. Developers integrate the technology through Python, C, and Java bindings to build private, offline voice interfaces without relying on proprietary cloud providers.

https://github.com/mozilla/DeepSpeech